Prologue

This week was dymaptic’s yearly company retreat. This year, we brought the whole team to Portland, Oregon, to work on improving ourselves so that we can deliver our best work for our clients. Our philosophy is that we need to take time to care for ourselves, to fill up our tank, so to speak, so that we can then take care of our clients.

But Christopher, this is an AI newsletter—you’re right, and I’m getting to it! There are actually lots of fun AI-related things that happened at the retreat (foreshadowing for future newsletters!), but I am running out of time to make sure this episode gets out this week, so I’m doing a simple one: Self-Driving Cars.

Cruiselogue

As part of our retreat each year, we have an “adventure day.” This year's adventures included going to a “low ropes course” to do some team building exercises, driving out to the Oregon coast, and then getting some ice cream at the Tillamook creamery.

What was different, though, was that I did some driving and rented a GIANT SUV to haul some of our staff around in—and it could drive itself*!

Levels of Car Driving Autonomy

Before we can talk about exactly how crazy and how fun this was, we need some common metrics to compare against—fortunately for self-driving cars, we have the levels of autonomy. Check out my friend Danny’s post for more details about these levels. Here is his quick summary of the levels:

• Level 0 = a totally normal old-school car
• Level 1 = old-school assistance like cruise control + lane-keeping assist
• Level 2 = self-driving BUT the human may have to yank back control if the car is about to kill you ← Tesla is here
• Level 3 = the human can read a book but must be ready to take over quickly if the car beeps ← Mercedes is dipping a toe into here
• Level 4 = no human in the driver’s seat, car safely stops if confused ← Waymo is here
• Level 5 = the AGI of self-driving, car handles anything a human can

What is “BlueCruise”

The first time it appeared on the dash, I thought it was related to Bluetooth because someone had [coincidentally] just connected their phone via Bluetooth to play some music. But as it turns out, on certain highways, in certain conditions, the car would drive itself without my hands on the steering wheel. I really want this to qualify as Level 3, but most sources seem to say it is only Level 2 because you are supposed to be paying attention to the road. My experience (paying little/less attention to the road) says that this was way more like Level 3, though.

It could automatically and safely change lanes, react to people cutting in front of it, and manage speed—all without my input. It couldn't navigate, so I had to take over to exit the highway. That is when BlueCruise was disabled, and I had to switch back to just the smart cruise control.

It did not tell me to monitor it, and in fact, the only time it complained was when I tried to take a video of it automatically changing lanes one evening. Whoops.

The Ford “BlueCruise” system only works on certain highways, but claims to cover 97% of all controlled access highways. It can’t help with city driving, but in a place like LA or Houston, where so much of the driving is on highways, it would make a huge difference!

In this frame from the video I tried to record of it changing lanes, you can see it yelling at me, but you can also see what is probably the driver-facing camera that I was blocking with my phone at the time.

Arguing with the AI Driver

The car also had fancy cruise control that could keep it in the lane and follow the road, but I still had to keep my hands on the steering wheel. This worked in almost all conditions; however, the sharp and windy corners of the Oregon coast were no match for it. It would often BEEEP at me, and then say “Turn too sharp” on the screen, often leaving me with very little time to react.

I think that is basically Level 2.

I spent most of the drive arguing, out loud, with the car—probably driving my coworkers a bit crazy. It never responded except to abandon me when the turns were too sharp, judge me for being too close to the centerline on a narrow road with a huge car, and fail to correctly identify where the lanes were when I turned it back on.

Maybe the driving AIs could learn something from the chatbot AIs. Whenever I submit something to the AI editors for feedback, they usually start with something like, "your draft is already quite strong!" Perhaps positive feedback would help drivers do better!

Better Than Tesla

I’ve driven several Teslas over the years, never with the full self-driving feature, though. This Ford was at least as good as the Teslas, and even better on the highway—no hands!

Despite arguing with it, the drive was much more relaxed for me. It required less constant input to the steering wheel, and although it reduced some of my tasks to just checking on it, it was really nice for this type of driving.

Although reviews were mixed in the car, with shock and amazement all around. Everyone settled in to see how the car would behave in traffic, especially on the return home.

I’m excited for the future where I don’t have to drive at all.

Newsologue

• Claude Haiku 4.5 comes out this week—A very cheap, very powerful model that will be great for powering free experiences on websites and other similar uses.
• The NVIDIA DGX Spark is now available, and I was on the early list, so now maybe I will buy one!
• MITs SEAL framework pushes us closer to self-improving models (I have previously written about why models feel like they are learning when they really are not.)

Epilogue

This post was put together at kind of the last minute, although that didn't cause me to lean on AI more. I wrote this post, Holly edited it. My friend Danny stood in my kitchen with me chatting about this before I finished writing as well.

No fancy AI tricks this week, but more soon!